login
terry-li-hm

chrome-automation

by @terry-li-hm in AI & LLM
0
0
# Install this skill:
npx skills add terry-li-hm/skills --skill "chrome-automation"

Install specific skill from multi-skill repository

# Description

Reference skill for Claude in Chrome browser automation best practices. Not user-invocable β€” use as internal guidance when automating Chrome.

# SKILL.md

name: chrome-automation
description: Reference skill for Claude in Chrome browser automation best practices. Not user-invocable β€” use as internal guidance when automating Chrome.
user_invocable: false
platform: claude-code
platform_note: Reference for Claude in Chrome MCP tools. OpenClaw has equivalent browser tools with different API.

Chrome Automation Best Practices

Reference for Claude in Chrome MCP tools. Consult when doing browser automation.

Session Startup

  1. Always create a new tab at session start using tabs_create_mcp
  2. Only use tabs you created in this session β€” never reuse tab IDs from previous sessions
  3. If a tool returns "tab doesn't exist", call tabs_context_mcp to get fresh tab IDs

Reading Page Content

  • read_page captures the full accessibility tree including content below the viewport
  • No scrolling needed to get below-fold content
  • Workflow: navigate β†’ wait 2 seconds β†’ read_page once
  • Use max_chars parameter if page is large (default 50000)

Window Sizing

  • Resize window before read_page to reduce tokens
  • 800x600 for chat apps (WhatsApp, messaging)
  • 1024x768 for general browsing
  • Large viewports waste context on empty space

Page Load

  • Always wait 2 seconds after navigation before reading
  • If content shows loading placeholders, wait longer or refresh
  • LinkedIn job pages may need extra time to hydrate

Common Gotchas

Issue Solution
Tab ID invalid Call tabs_context_mcp to refresh
Content not loading Wait longer, or navigate directly to URL
Screenshot from wrong tab Verify tabId in tool response matches intended tab
LinkedIn "Saved" toggles Clicking "Saved" unsaves β€” use three-dot menu for actions
WhatsApp message direction Left/white = incoming, right/green = outgoing
Gmail contenteditable form_input unreliable on Gmail compose β€” write draft to file instead

Tool Selection

Task Tool
Get page structure/content read_page
Extract article text get_page_text
Click elements computer with ref parameter
Type text computer action=type
Navigate navigate
Screenshot for debugging computer action=screenshot
Create new tab tabs_create_mcp
List available tabs tabs_context_mcp

Login-Required Sites

These sites return login walls via web scraping tools β€” must use browser automation:
- LinkedIn
- X/Twitter
- WhatsApp Web
- Gmail (for some operations)

Alerts and Dialogs

Avoid triggering JavaScript alerts, confirms, or prompts β€” they block all further browser events. If you must interact with dialog-triggering elements, warn the user first.

Session Cleanup

Always navigate to the idle page after completing browser automation tasks:

https://terry-li-hm.github.io/claude-home/

This keeps the browser in a clean, non-distracting state instead of leaving it on a logged-in dashboard or sensitive page. The idle page shows a minimal dark background with subtle day/time in the corner.

  • /evaluate-job β€” Uses Chrome for LinkedIn job extraction
  • /review-saved-jobs β€” Batch processes LinkedIn saved jobs

# Related Skills

# Supported AI Coding Agents

This skill is compatible with the SKILL.md standard and works with all major AI coding agents:

⚑ Amp πŸš€ Antigravity πŸ€– Claude Code πŸ¦€ Clawdbot πŸ“ Codex ▢️ Cursor πŸ€– Droid πŸ’Ž Gemini CLI πŸ™ GitHub Copilot πŸͺΏ Goose πŸ“Š Kilo Code πŸ”§ Kiro CLI πŸ’» OpenCode 🦘 Roo Code 🌲 Trae πŸ„ Windsurf

Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.